Alvin Liu (MD) sat down with Sheryl Stevenson, Group Editorial Director of Ophthalmology Times®, to discuss his presentation on Deep Learning and 3D OCT at the ASCRS Annual Meeting in San Diego.
video transcript
Editor’s Note: This transcript has been edited for clarity.
Cheryl Stevenson:
I joined Dr. Alvin Liu, who will be presenting at ASCRS this year. welcome. Tell us a little more about your presentation on deep learning and 3D OCT.
Alvin Liu, M.D.:
Cheryl, thank you for speaking with us today. I am happy to share the results. Please allow me to introduce myself a little more. My name is Alvin Liu. I am a retina specialist at the Wilmer Eye Institute at Johns Hopkins University.
My research focuses on the application of artificial intelligence to ophthalmology. In particular, I am also Director of the Wilmar Precision Eye Center of Excellence. Therefore, the work she presents at ASCRS this year is directly related to our Center of Excellence.
The overall premise is that we know that macular degeneration is the leading cause of central vision loss in older adults in the United States and around the world. I lose my sight because of my mold. Especially in the case of wet AMD, early and timely treatment to improve current vision predicts improved eventual vision. Therefore, it is imperative to identify patients at high risk for imminent transition to wet AMD.
There are now methods that use ERAS criteria to provide average estimates of conversion or progression to advanced AMD. However, the ERAS criteria can only provide average risk estimates over 5 years. The model we developed can be used as a tool that can provide information in a more reasonable or more meaningful period of 6 months. We start by asking ourselves whether deep learning, the state-of-the-art artificial intelligence technique, can be used for medical image analysis. Using deep learning, could he analyze OCT images and predict an impending conversion from dry AMD to wet AMD within six months?
To do so, we collected a dataset of over 2500 AMD patients and over 30,000 OCT images. Using only OCT images, we train a model that can generate robust predictions when an eye is at high risk of transforming to wet AMD within 6 months. In addition, we conducted various experiments to see what would happen if we gave this model additional information in the form of a number of clinical variables available, such as patient age, gender, visual acuity, and eye condition. Demonstrating that in anticipation of an imminent conversion to Wet AMD. This additional tabular clinical information was also helpful, as it meant the patient’s first eye, i.e., a patient who had never had exudative AMD in either eye.
